The smallest thing that prices a job properly.

MVP specification

Platform: responsive web app built on Hono + Cloudflare Pages. The engine is a pure TypeScript module shared by the client bundle and the server routes, so a quote is always reproducible. Storage: Cloudflare D1 (SQLite) keeps the business profile, overhead lines and saved quotes. Calculations run entirely on the device, so the app keeps working offline.

User journey

First run (target: first quote in under 4 minutes)

On-site quote (core loop)

Follow-up and habit

Wireframes

Mobile UI design

Database / data structure

-- Cloudflare D1 (SQLite). All money stored in cents or REAL, AUD. All prices stored ex-GST + gst separately.
businesses     (id, name, abn, phone, email, suburb, state, gst_registered,
                employee_rate, owner_rate, sub_rate, supplies_per_job, travel_per_km,
                target_margin, billable_hours_pm, jobs_pm, terms, updated_at)
overhead_items (id, business_id, label, monthly_amount, sort)
settings       (id PK, red_below_pct, green_at_pct, round_to, overhead_method, basis)
quotes         (id, customer, suburb, status, job_type, created_at,
                inputs         TEXT,   -- JSON snapshot: property, labour, costs, overhead, target
                engine_version TEXT,   -- e.g. "1.0.0" — results reproducible forever
                chosen_price_inc REAL, message_tone, message_text,
                parent_quote_id TEXT)  -- duplicates

Storing the input snapshot plus engine version means a quote always shows the numbers it was sent with, even after defaults or the engine change. The client also caches state in localStorage so the calculator keeps working offline and syncs when reconnected.

Calculation engine specification

Definitions (strict)

Solving for the price

Let C = Cd + O_fixed when the target is after overhead (the default), or C = Cd when it's on direct costs only (classic gross margin). Let p be the %-of-revenue overhead (0 if another method is used).

Gross margin M :  P = C ÷ (1 − M − p)          // NOT C × (1 + M)
Markup K       :  P = C × (1 + K) ÷ (1 − p(1 + K))
Profit $T      :  P = (C + T) ÷ (1 − p)
Guard          :  if denominator ≤ 0 → error "target + overhead % must be < 100%"
Rounding       :  P rounded to cents first, then GST = round₂(P × 0.10)

Rounding options

For step s (default $5): offer floor, ceil and ceil + s of the inc-GST price. Each is re-evaluated: P' = price ÷ 1.1, then margin, profit and status are recalculated. The recommended option is the lowest round-up, never the cheapest. Options below the calculated price are marked as trimming margin.

Status

Red if profit after overhead is below 0 or the margin (on the chosen basis) is under the red threshold (default 15%). Amber if it's under the green threshold (default 30%). Otherwise green. Both thresholds are configurable.

Time suggestion (not a price input)

Transparent rule-of-thumb: minutes per room type × job-type multiplier (bond 2.1×, deep 1.6×, builders 2.6×…) × condition (0.85–1.75) × frequency (regular only) × storeys × pets + access. Commercial and builders use floor-area rates. The user must accept it; it never changes hours on its own.

Example calculations engine self-test: pass

Computed live below by the same engine.ts that runs the app.

A · The brief’s test case: 30% gross margin on direct cost, before overhead

✓ PASS: expected $292.86 + $29.29 = $322.15

B · Same job, target after overhead (per labour hour, $1,450/mo ÷ 320 h)

C · Owner-operator doing the job solo. Owner time valued at $45/h

D · The markup trap: 30% markup vs 30% margin

E · Bond clean with % of revenue overhead (12%) and $80 profit target

F · Not GST registered

AI assistant rules

• Explanation first, AI second. The “Why this price?” paragraph is generated deterministically from engine output, so it can never be wrong. The AI only answers follow-up questions.
• The AI prompt contains only the calculated figures (no customer name), and is told to use only those, never invent market or competitor prices or wage rates, and never give tax, legal or accounting advice.
• “Polish message” may rephrase only. The result is rejected if the price string disappears; the leak-check still runs.
• Every AI output is labelled and user-editable. Nothing is sent automatically. The API key never reaches the client — all model calls go through the server route.

Market price comparison

Stored with a confidence level, source and retrieval date, at Australia, state or city level. Displayed as a bar with a “You” marker. It never changes the calculated price. Current data status:

Before launch: collect 30+ published price lists per capital city (bond, regular hourly, Airbnb turnover, commercial hourly, builders per m²) and record each source.

Job types and multipliers